Variant Calling outside Torrent Suite and TVC

[gmarco]

Hello,

After trying to build Torrent Suite 4 for weeks on a RedHat Linux without success (I got compiled programs but not working with my Linux Architecture), and exploring all the tools they "offer" inside their analysis suite.

I'm scared.

They use their own variant caller (tvc), a tool that supports hotspots.. freebayes and normalized measurements. In my opinion an in-house variant caller to deal with the homopolymer problem that we all know.

They also use a modified version of GATK, that includes the IndelAssembly toolkit. I think it's their own modified version of GATK. I can't find any information about IndelAssembly (toolkit intended to find large indels) outside Life realms.

I would like to analyze some Ion Torrent data, but I don't like this way. I would like to know what tools are you using for variant calling outside the straight way from Life and they Torrent Suite one click analyzer. Tools supported by all the scientific community. For me a golden standard version of GATK would be ok.

My purpose at this moment is to analyze AmpliSeq exomes.

Thanks !

[TiborNagy]

If you analyse human data, GATK is a very good tool.

[gmarco]

I'm using GATK as I do with Illumina data. I'm experiencing very slow variant calling process with UnifiedGenotyper.

[snetmcom]

Could you ask your service provider to analyze the data on their Torrent Suite? It extremely simple on the Server.

[c_ro87]

Hello, i'm analyzing Ampliseq runs on custom gene panels, using a 316 Chip

i also want to try something different, i also try the GATK, but after the MarkDuplicates step, i found that due to my amplicons start and end in the same locations in the alignment thay are marked as PCR duplicates and removed from the next steps..

so i can't do this step, if i follow the best practices guidelines, and aplly hard filtering, i get a lot more variants that with the ION variantCaller pipeline

i mean aprox 30 variants in the ION pipeline for each barcado, and aprox 150 with GATK after the hard filtering step

i don't know how good is this disagrement

what experience do you have?

[snetmcom]

Ampliseq library will be mostly duplicates by design. You should not be filtering these reads out.

GATK is likely calling false positives because it does not have any ion specific rules.

[arnaud83]

I'm glad to see people who think like me about the Torrent Server Suite and its Tool Box.
I'm used to deal with PGM data, and here is my pipeline :

-Alignment is done with bwasw. I tested several alignment programs (tmap, novoalign) and even if the percentage of mapped reads is 4-5 % smaller than tmap, the indel mismatch is better.

-For an exome i use MarkDuplicates.jar from PICARD. For targeted sequencing, it's not recommended because 80-90% of reads will be marked.

-Then, i use FreeBayes to call SNP and UnifiedGenotyper to call INDEL. I prefer to use FreeBayes to call SNP because i can set the min variant frequency and be more sensitive than UnifiedGenotyper. But It requires additional filtering (Strand Bias, Quality, etc...).
For more specificity without to much work, i suggest you UnifiedGenotyper for both SNP and INDEL.
But to be honest you have better chance to call a true INDEL with flipping a coin. There are too many False Positive, and for an exome it's a misery.

[c_ro87]

@arnaud83: how you do the strand bias filtering?

[arnaud83]

@arnaud83: how you do the strand bias filtering?

For strand bias based on Fisher's exact test (Unifiedgenotyper), i use a threshold of 60 ( p=0.000001). Variants below this threshold will be keep.
FreeBayes doesn't include strand bias in the vcf output, but you can easily compute this with some programming skills

[IonTom]

I also kind of gave up on the IonTorrent Suite.

Currently i am using NextGenMap for Alignment.

For Variant Calling i use Platypus. The principle is kind of similar to FreeBayes,
but the QC statistics and filters are much more complete. As in everything you can think of.

One additional fillter I use is implemented in the BioConductor VariantTools package.
It tells you at how many different in read positions a variant was found.
This is kind of important as removing PCR duplicates is not really possible for
Amplicon data.

[arnaud83]

I did not know these tools. Thank you.
I will test them.

[IonTom]

@arnaud83: How did they work for you ?


There is a nice paper discussing the topic of using aligners on ion torrent data:

http://www.biomedcentral.com/1471-2164/15/264/

[wolfpack14]

The homopolymer issue in IonTorrent can be semi-mitigated through setting frequency thresholds based on mixture fractions. The solution can be applied through post-processing or integrated into one of these variant caller applications. We're working on a paper right now that demonstrates the methodology in a productional lab environment (vs. academic environment you see in most papers).

[arnaud83]

@arnaud83: How did they work for you ?


There is a nice paper discussing the topic of using aligners on ion torrent data:
http://www.biomedcentral.com/1471-2164/15/264/

Well, to be honest, i'm a little bit disappointed by mosaik. The mentioned paper shows promising results but i obtained worse results than bwa or tmap.

[gmarco]

I'm very happy seeing this topic has received many answers. I'm wiling to try all these tools.

The homopolymer issue in IonTorrent can be semi-mitigated through setting frequency thresholds based on mixture fractions. The solution can be applied through post-processing or integrated into one of these variant caller applications. We're working on a paper right now that demonstrates the methodology in a productional lab environment (vs. academic environment you see in most papers).

Hello wolfpack do you have any ETA?

I expected very very very slow GATK UnifiedGenotyper variant calling with Ion Torrent exome variant calling. Anyone had this issue?

Ion Torrent data has 2 major issues:
1 - Dealing with homopolymer problem (how the hell we're supposed to filter those reads, or deal with them)
2 - Setup correct variant calling settings.